Abstract:
Stitching multiple images is a requirement for a wide variety of applications including medical image analysis, panorama creation, high resolution video content generation, immersive virtual environments. The quality of stitching is measured visually by observing the similarity of the stitched image to each corresponding input image and by the alignment of structure in the stitched images. Feature based image alignment algorithms are used for structural alignment but they may not be able to align structure properly. To overcome this shortcoming structure deformation based algorithms have been proposed. Since these algorithms are based upon optimum seam calculation for finding regions that will perfectly align structure they are used only for still images. Extending these algorithms to stitching of multiple images, to form a higher resolution video, results in a very unpleasant seam movement artifact. To resolve this problem two methods have been proposed in this thesis: Average Seam Method and Interpolated Seam Method.
In the first method, Average Seam, we calculate an average seam for fixed number of frames, n, and stitched n frames using this average seam. This method reduces the frequent appearance of the seam jumping artifact but could not overcome it completely. In the second method, Interpolated Seam, we find three seams in every mth and rest of the seams calculated by the interpolation of these seams. This method not only successfully overcome the seam jumping artifact in the resulted video but also decrease the computational time significantly
